Projects

Developed an image captioning model using deep learning, integrating InceptionV3 for image feature extraction and an LSTM-based network for text generation. Applied NLP techniques to enhance textual representation. Optimized training with bidirectional LSTMs, batch normalization, and dropout. Achieved a BLEU-4 score of 25 on the Flickr8k dataset.

Led the Business Data Management project, to transform the sales and inventory strategies for a fertilizer business. This project encompassed detailed primary data collection, rigorous data analysis, and formulating actionable strategies to drive business growth. 

• Conducted in-depth data analysis of customer buying patterns to identify top-selling products and seasonal preferences.

• Developed targeted marketing strategies by analyzing sales data from top purchasing villages and tailoring campaigns to local preferences.

• Implemented seasonal analysis to ensure product availability during peak demand periods, enhancing customer satisfaction.

• Addressed frequent stockouts by calculating precise reorder points for each product.

• Leveraged statistical techniques to inform strategic business decisions, resulting in a 5% increase in sales and a 10% reduction in excess inventory over the next quarter.

This project involved an extensive Exploratory Data Analysis (EDA) on a loan defaulter dataset, and developing a robust machine learning classification model to predict loan defaulters using an extensive dataset. 

• Data Preprocessing and Feature Engineering: Cleaned the dataset, reduced dimensionality, and extracted key features such as income-to-loan ratio, credit history length, etc, to enhance the model's predictive power.

• Implemented Logistic Regression, Random Forest, and Gradient Boosting algorithms, achieving 89% accuracy and an F1-score of 0.85.

• Demographic Insights: Identified that females, married individuals with fewer children, & those with stable housing are less likely to default, while high-income individuals with loans close to their income are a higher risks.

• Risk Segmentation: Developed a detailed risk segmentation model, identifying low-risk segments and high-risk segments, enabling more targeted and safer lending practices.

• Strategic Impact: Provided data-driven recommendations that can reduce loan default rates, saving institutions significant financial resources.

Accurate sales forecasts enable businesses to meet customer demand effectively while minimizing costs associated with stock issues. This project leverages machine learning techniques to predict sales with the finest accuracy, offering profound insights that drive strategic business decisions.

• Handled missing values and outliers to ensure data quality. Conducted feature engineering to create meaningful variables for analysis.

• Revealed key sales drivers, including holidays, promotions, and store types, with visual insights from advanced plots. Identified seasonal trends, sales patterns, contributing to a deeper understanding of market dynamics.

• Implemented multiple regression models, and The Random Forest model emerged as the top performer, Achieving an astounding 93% prediction accuracy, maintaining the industry standards.

• Identified actionable insights into factors driving sales, aiding in strategic planning. Facilitated targeted marketing efforts by identifying key periods and products.

This project focuses on building a machine learning model to predict taxi fares in New York City, leveraging a variety of data science techniques. By handling real-world challenges such as missing data, outliers, and feature engineering, the project showcases a comprehensive approach to creating robust predictive models suitable for deployment in dynamic environments.

Detailed Project Description:

• Data Cleaning: Addressed missing values across key features like passenger count and rate codes using imputation techniques, resolved anomalies such as negative fare amounts, and inconsistent timestamps.

• Outlier Handling: Identified and managed outliers in numerical columns such as trip distance and total amount using statistical methods like interquartile range (IQR) and feature transformations.

• Feature Engineering: Created new features including trip duration (calculated from pickup and drop-off times) and a late-night ride indicator to capture patterns unique to certain times of day.

• Categorical Feature Encoding: Applied one-hot encoding to categorical variables like payment type and store-and-forward flags, enabling effective use in machine learning models.

• Data Transformation: Normalized skewed distributions of critical features such as trip distance using logarithmic transformations to improve model performance and reduce the impact of outliers.

• Exploratory Data Analysis (EDA): Conducted visual and statistical analysis to identify correlations, trends, and feature importance, providing insights into the factors influencing taxi fares.

• Model Development: Evaluated multiple regression models, including Linear Regression, Decision Trees, Random Forests, Gradient Boosting, and XGBoost, to identify the best-performing approach.

• Hyperparameter Tuning: Leveraged GridSearchCV to optimize model parameters like tree depth, learning rate, and number of estimators, significantly improving accuracy and generalization.

• Performance Metrics: Assessed model performance using Mean Squared Error (MSE) and R² scores, achieving a high R² on validation datasets while minimizing overfitting.

• Final Deployment: Built an automated preprocessing pipeline to handle cleaning, transformation, and feature extraction, ensuring a streamlined workflow for future data predictions.

Developed a robust multi-user library management system with two roles: Librarian and General User, enabling role-based access and functionalities.

• Librarian Features: Librarians can manage users, create and manage book sections, upload e-books to Google Drive, approve/revoke access to books, and generate monthly activity reports with Celery and Redis for automation.

• User Features: Users can request up to 5 books, rate and review books, organize books into personalized shelves, update their profiles, and download e-books after purchase with lifetime access.

• Google Drive Integration: Integrated Google Drive API to store and manage e-book PDFs and cover images, with automated permissions for user access control and viewer restrictions.

• Backend Architecture: Built with Flask for backend APIs, JWT for authentication, and SQLite for data storage. Used SQLAlchemy for ORM, enabling efficient database management and interactions.

• Frontend Development: Designed an interactive and responsive user interface using Vue.js, providing a seamless experience for both librarians and general users.

• Task Automation: Leveraged Celery and Redis for background tasks such as sending monthly reports, daily reminders, and managing book access permissions automatically.

• Real-time Data Handling: Utilized Redis for caching APIs to enhance performance and ensure faster data retrieval for frequent user actions.

• API Design and Security: Developed secure RESTful APIs with Flask and JWT, ensuring safe user authentication and authorization, including role-based access control for librarians and users.

• Comprehensive Reports: Built APIs for generating detailed user activity stats, and book request history for librarians, with Jinja2 templates for monthly activity report generation.

Tickets Booking Application is a multi-user application designed with modern web development frameworks. It provides distinct user and admin portals, allowing users to effortlessly book tickets and enabling admins to manage venues and shows with ease. 

• Built with Flask for robust backend development, Jinja2 templates for dynamic HTML generation, and Bootstrap for responsive design. Utilizes SQLite for efficient and reliable data storage.

• Users can view show details, select the number of tickets, and complete their booking. The system displays all booked tickets in a dedicated bookings page, ensuring users can easily keep track of their bookings.

• Admins can create new venues and shows, edit existing ones, and remove them as needed, with all changes reflected immediately in user interface.

• Users can search for shows based on location, tags, and ratings, making it simple to find the most suitable options.